UCD GEARY INSTITUTE FOR PUBLIC POLICY
DISCUSSION PAPER SERIES
Economic uncertainty, parental selection and
children’s educational outcomes
Arnaud Chevalier
IZA
Royal Holloway, University of London
UCD Geary Institute for Public Policy, University College Dublin
ROA, Maastricht University
SFI, Copenhagen
Olivier Marie
ROA, Maastricht University
CEP, London School of Economics
IZA
CESIfo, Munich
Geary WP2015/06
April, 2015
UCD Geary Institute Discussion Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. A revised version may be available directly from
the author.
Any opinions expressed here are those of the author(s) and not those of UCD Geary Institute. Research published in
this series may include views on policy, but the institute itself takes no institutional policy positions.
Economic Uncertainty, Parental Selection,
and Children’s Educational Outcomes*
Arnaud Chevalier• and Olivier Marie+
Abstract: After the fall of the Berlin Wall, East Germany experienced an unprecedented
temporary drop in fertility driven by economic uncertainty. Using various educational
measures, we show that the children born during this nativity slump perform worse
from an early age onwards. Consistent with negative selection, mothers who gave birth
in that period had worse observed personal characteristics. These children are also less
likely to have grown up within stable family environment. Investigating underlying
mechanisms reveals that parental educational input and emotional attachment were
also lower for these children. Finally, sibling analysis enable us to reject time of birth
effects.
Keywords: Parental selection; fertility; economic uncertainty; education
* We thank Derek Neal and two anonymous referees for their comments on an earlier version of
the manuscript. We are also grateful for feedback on previous versions of the paper from Rajeev
Dehehjia, Thomas Dohmen, Nicola Fuchs-Schündeln, Gauthier Lanot, João de Mello, Christian
Traxler, Tanya Wilson, Justin Wolfers and participants at various workshops, conferences and
seminars. Note that part of the text is similar to an earlier version of this work looking at a
different outcome. This version supersedes the previous one. We also want to thank IQB –Berlin
for providing us access to the IGLU2001 and PISA German Oversamples datasets, as well as DJI
and DIW for granting us access to the DJI survey and SOEP, respectively. Chevalier would like
thank the British Academy (MD133047) and GSBE (M.11.5122) for grants that made this
research possible. Marie is grateful for financial support for this research from the Netherlands
Organization for Scientific Research (NWO Veni # 451.12.005).
•
IZA, Royal Holloway, Geary Institute, ROA, and SFI. Email: [email protected]
Maastricht University, ROA, CEP, IZA and CESIfo. Corresponding Author: Department of Economics,
Maastricht University, P.O. Box 616, 6200 MD Maastricht, the Netherlands. Tel: +31 43 388 4605 Email:
o.marie@maastrichtuniveristy,nl
+
1
1. Introduction
This paper documents how the socio-economic environment not only affects the
size but also the composition of a cohort. Using a natural experiment, we show how
cohorts born during a period of high economic turmoil perform worse on various
dimensions of education. We subsequently study the possible mechanisms, focusing on
family composition and parenting behavior that could lead to negative effects on
education. The results confirm the large effect of parental selection. Further tests
dismiss alternative explanations for the worse educational attainment of the affected
cohorts.
Becker (1960) and Ben Porah (1973) have long hypothesized that fertility is a
pro-cyclical decision, see Lindo (2010) or Schaller (2012) for recent empirical evidence.
Gronau’s (1977) model suggests that an economic slump results in a negative income
effect, which reduces the demand for children; moreover, since children require a large
parental time investment, it also prompts a positive substitution effect that pushes the
demand for children in the opposite direction. Which effect is stronger is a priori
ambiguous but since fertility is pro-cyclical, the income effect appears to dominate
overall. However, the relative size of the income and substitution effects may differ
across family types, leading to the economic environment affecting the size of a cohort,
as well as its composition. Using education as a proxy for earning potential, Perry (2004)
argues that for completed fertility, the income effect dominates for high education
women, while the substitution effect dominates for less educated ones. If this were also
true for short-run variations in income, then cohort composition would be pro-cyclical.
Indeed, Dehejia and Lleras-Muney (2004) show that white mothers giving birth when
unemployment is higher are less educated, resulting in worse health outcomes at birth,
in a mechanism that Currie, Duque and Garfinkel (2015) confirm was also at play during
the latest economic recession in the US. Del Bono, Weber and Winter-Ebmer (2014) also
highlight that the fertility drop associated with another type of economic shock, namely
plant closures, is solely driven by more skilled women postponing pregnancies.
This paper improves on the existing literature in three important dimensions.
First, we rely on much larger variations in the economic environment, which were
largely unexpected. More precisely, following the fall of the Berlin Wall and the collapse
of the collectivist socio-economic system, East Germany went through a period of high
economic uncertainty. Concomitantly, the fertility rate in the former GDR was more than
2
halved over a three-year period, an unprecedented peace-time event, before stabilizing1.
Throughout the manuscript, we refer to these cohorts born in the Eastern Länder2
between August 1990 and December 1993 as the ‘Children of the Wall’ (CoW). The
natural experiment that we exploit led to a very profound yet short-lived exogenous
fertility shock in the former East Germany, which creates clear pre- and post-cohorts.
Moreover, no drop in fertility was observed in the former West Germany, which
generates a natural control group since those born on either side of the “border” were
subject to increasingly similar socio-economic environments when growing up in reunified Germany. Also, note that by the time these children entered schools, disruption
to the school system stemming from the end of communism had largely subdued. Indeed
we reject that for the cohorts and outcomes of interest, trends in East and West
Germany substantially differ. This natural control group enables us to credibly account
for the potential effect of shared macro shocks and we thus apply a difference-indifferences estimator strategy throughout. Some of the analysis is also conducted within
school cohorts amongst children conceived only a few months apart around the fall of
the Wall, as a way to ensure that the economic or school environment do not directly
drive our results.
Second, this paper investigates the longer run consequences of parental selection.
Recent research has highlighted the importance of endowment, early conditions and
parental investments on the accumulation of human capital (Cunha and Heckman,
[2007] or Bjorklund and Salvanes [2010] for reviews). In particular, Cunha, Heckman,
Lochner and Masterov (2006) show the high returns to early investment. As such, one
may expect that changes in parental selection lead to differences in the accumulation of
human capital between cohorts. Using four different datasets, we are able to document
variations in educational attainment from age 10 to 17. Since the cohorts of interest are
much smaller than usual, we can immediately reject any crowding out effect and, by
contrast, we would expect these cohorts to have experienced higher level of public
spending; indeed, class size dropped by 10% for the affected cohorts (Kempkes, 2010).
1
Other East-European countries also experienced drops in fertility following the collapse of the communist
regimes in place, although their magnitudes were substantially smaller than that observed in East Germany
(UNECE, [2000]).
2
Throughout the paper, we will use “Land” or “State” interchangeably to refer to the 16 constituent states of the
Federal Republic of Germany, or mostly 15 as Berlin is often dropped from the analysis since it is not possible to
know which individuals were from the East or West after unification. Note also that the plural of Land is Länder.
3
Consequently, if parental selection proves negative for these children, our results should
be interpreted as lower bound estimates of the true effect of parental selection.
Third, the literature on parental selection has been mostly unable to
comprehensively document certain parental characteristics that are likely to be
associated with both fertility selection and children’s outcomes. We fill this gap by
exploiting very rich individual level datasets with information on mother and child
characteristics. We begin our analysis by considering the commonly used observable
maternal characteristics (age, education and employment/income) to establish the
direction of the selection into fertility. This data also enables us to expand on several
previously mostly overlooked sets of characteristics: i) family structure3; ii) parental
input in the child’s education; and iii) maternal emotional attachment and parenting
competence as expressed by the children themselves. As such, we more precisely
document the parental selection and assess potential mechanisms by which it affects
children’s outcomes. We also provide the first direct micro-evidence that uncertainty
about the economy affects a woman’s fertility decisions differently by education level.
A remaining worry would be that children born during this very uncertain time
suffered from the possible adverse economic environment that they and their mothers
faced. For example, the fetal programming hypothesis (Barker, 1995) asserts that
parental stress while in the womb can lead to abnormal emotional control (see van den
Bergh et al. [2005] for a review), which could in turn increase the chances of negative
outcomes, even without parental selection (see Aizer, Stroud and Buka [2012], for
example). Our first examination of this mechanism uses school test data comparing
children from the same school cohort born nine months around the cut-off defined by
the fall of the Berlin Wall. These children can be considered to have shared the same
economic environment during their childhood and the same educational environment in
the year of the test; as such, they only differ according to parental selection. Nonetheless,
certain mothers may have experienced some level of stress while pregnant, albeit at
different months of the pregnancy, or during the early months of childrearing. Since the
timing of the stress might be important, we also conduct a second test that relies on
comparing CoW with their older siblings born in the economically stable times of East
Germany. These older siblings are mostly expected to display similar outcomes if these
3
Neal (2004) highlights family structure as an important long run selection mechanism that explains the large
female black-white income gap in the US, albeit not in the specific context of an economic recession.
4
are driven by parental selection, but not if our reduced form evidence is driven by being
born in a particular environment.
Our main empirical analysis and the ensuing findings developed in the paper are
as follows. We first clearly document the unprecedented drop in birth rate observed in
East Germany just after the fall of the Berlin Wall and especially the drop in in-wedlock
birth. We subsequently offer a number of explanations as to why it happened in the
context of the historical and institutional background, whereby the very high level of
economic uncertainty in East Germany in the years following German reunification
emerges as the main reason for the fertility reduction. In the absence of any national
test, we exploit German oversamples at two international tests (Progress in
International Reading Literacy Study or PIRLS and Program for International Student
Assessment or PISA) to objectively assess the performance of CoW at ages 10 and 15,
compared to their classmates. Additionally, we use the “Deutsches Jugend Institut”
Youth Survey (DJI) and the German Socio-Economic Panel (SOEP) to assess self-reported
educational outcomes at ages 12 and 17. The results are consistent across datasets and
highlight that the affected cohorts experienced worse educational outcomes at all ages.
These findings are confirmed in a number of alternative specifications and placebo
checks enable us to reject the notion that the results are driven by time-specific
unobservable characteristics.
Having documented the worse outcomes of the CoW, we investigate the selection
into parenthood of the mothers of these children. Using our various sources, we report
strong evidence of the negative selection of women who gave birth in East Germany just
after the end of the communist regime. On average, these women were younger, had
lower levels of education, were less likely to be economically active and were more often
on welfare. From the SOEP and DJI, we also document that CoW grew up in
environments with much less stable family structures. The effect is unlikely to come
from poverty as families with children born after the fall are not poorer, thanks to the
relatively large social transfers. The various datasets allow us to further explore the
possible mechanisms that could explain the worse educational outcomes of CoW,
whereby we uncover substantial differences in parental behavior; for instance, these
parents were less likely to read to their children and more generally provided less
educational inputs. Importantly, CoW self-report their relationship with their families as
being of much lower quality. This lower emotional attachment is often put forward in
5
the child development literature on the long-term effect of early rearing conditions (see
Brook-Gunn, Berlin and Fuligni [2010] or Conti and Heckman [2013] for a recent
review).
We reject the hypothesis that these children have worse outcomes due to being
born in bad economic times. First, such an explanation would not involve differences in
parents’ characteristics and behavior. Second, the test score analysis involves comparing
children in the same school born only a few months apart, as such environmental
differences cannot explain the large differences in educational attainment. While one
could still agree that the timing of the stress while in the womb matters, the CoWs’ older
siblings - who were born under the stable times of the communist regime - also had
worse educational outcomes and poor relationships with their mothers. A final test to
confirm that the nature of the parental selection mechanism observed is driven by
economic uncertainty is proposed. We combine individual information on a women’s
education level, her believes about future economic conditions and her fertility in the
years after the fall of the Berlin Wall. This enables us to prove that those who were more
worried about the future were less likely to have children, but also that crucially this
response was much stronger for women with higher levels of education.
Our findings clearly show that the cohort of children born during the period of
substantial economic uncertainty that we study was negatively selected. This conclusion
has potentially important policy implications. First, the provision of public services (e.g.
school investment) should not only be based on the size of an incoming cohort; rather,
more attention should be paid on its composition. Second, since remedial policies are
most effective when taking place at an early age (Cunha and Heckman [2007]), it is
important to identify at-risk children as early as possible, and even perhaps before they
are born. Through targeted policies such as home improvement programs, improving
those skills could thus also have a large impact on negatively selected cohorts (see
reviews of evidence in Doyle et al, 2013). However, our findings also suggest that it
might be difficult to identify the right target group of parent/children since the selection
is driven by characteristics such as parenting skills or emotional attachment that are
typically not observed.
The remainder of this paper is structured as follows. Section 2 details the
institutional background surrounding the period of the fertility drop that we exploit as a
natural experiment and considers various possible explanations on why the fertility
6
dropped. Section 3 describes the various datasets used and specifies the differences-indifferences strategy that we adopt throughout. Section 4 presents evidence on the
differences in educational outcomes for the CoW compared to other cohorts. Section 5
documents the extent of parental selection and the differences in parental behavior,
before Section 6 offers concluding remarks.
2. Documenting the Fertility Drop
2.1 East Germany and the German reunification
Germany was split along the positions of the occupying armies in the aftermath of
World War II, with the Federal Republic of Germany (FRG or West Germany) and the
German Democratic Republic (GDR or East Germany) being officially founded in 1949.
The GDR developed as one of the most orthodox of the former European Communist
regimes. As the two countries’ economic and political performances diverged,
increasingly more citizens from East Germany migrated by crossing the border into
West Berlin. To stop this exodus, a wall was built around the western part of the city in
1961, with the Berlin Wall becoming the symbol of the forty-year physical and socioeconomic separation of people who had previously shared a common destiny.
By the end of the 1980s, a series of sudden and radical political changes led to the
rapid collapse of the communist regimes in most of Eastern Europe. In the GDR, large
demonstrations against the regime started in September 1989 and emblematically
culminated with the televised demolition of the Berlin Wall on the evening of 9th
November 1989, as the borders between East and West Germany were declared opened.
There was a strong political will to quickly re-unite the two countries, especially with
the organization of election as early as March 1990 where the communist party in East
Germany was heavily defeated. Two month later, a common currency was announced
and introduced in July of the same year. Unification was officially completed in October
1990, less than 11 months after the Berlin Wall had fallen (see for example, Judt [2005]
for details). The very abrupt end of almost half a century of communist rule and the
express reunification that followed was a huge unexpected shock, leading to a period of
great socio-economic uncertainties for citizens of the new East German Länder4. This
4
We are not the first to use German re-unification as a natural experiment to investigate the occupational effect
on precautionary (Fuchs-Schündeln and Schündeln [2005]) and household saving (Fuchs-Schündel [2008]),
7
was perhaps best illustrated by the unprecedented decline in the number of births that
occurred there in the years immediately after the fall of the Wall.
2.2 The Fertility Drop
The upper panel of Figure 1 documents the yearly crude fertility rate in East and
West Germany from 1950 to 2008. The first thing to note is that despite the somewhat
lower level in the East, the trends in fertility up to 1989 were very similar in both
countries: a post-war baby boom until the mid-1960s, a rapid decrease (readjustment)
in the following decade, before a relative stabilization between 1970 and 1990. The
somewhat larger increase in fertility in East Germany starting in 1974 was the result of
pro-natalist policies that provided a range of welfare benefits to parents (see Reinheckel
et al. [1998] for details). However, these policies only had a temporary effect so that
fertility trends in both countries were similar by the mid-1980s5, which is the origin of
the study period. What stands out in Figure 1 is the massive and temporary collapse in
birth rates in the East, but not in the West, following the fall of the Berlin Wall (vertical
red line). It has been defined by demographers as the “most substantial fall in birth rates
that ever occurred in peacetime” (Conrad, Lechner and Werner [1996], p. 331). Within a
year, the birth rate dropped by 40 percent, before reaching an all-time low in 1993,
when it was less than half of its 1989 level. However, this fertility drop was relatively
short lived and a recovery started in 1994.
[Figure 1 about here]
All family structure decisions were affected by the regime change. The lower
panel shows the yearly marriage rate (per 1,000 inhabitants) in East and West Germany
over the same period. In the 1950s and 1960s, the rates are remarkably similar in the
two countries, declining from 10/1,000 to 7.5/1,000. However, pro-natalist policies
introduced in East Germany in the early-1970s (see Reinheckel et al. [1998] for details)
temporarily pushed the marriage rate up, whereby the marriage rate in East Germany
preference for redistribution (Alesina and Fuchs-Schündeln [2007]), consumption behaviour (Bursztyn and
Cantoni [2012]) or the economic impact of networks (Burchardi and Hassan [2013]). No study has however
previously focused on the outcome of the children born during this period as we do in this paper.
5
Note that the cohort of women coming to their peak fertility age after 1989 was relatively smaller, having been
born during the fertility ebb of the early-1970s. This natural cohort size effect contributes at most to 10 percent
of the drop in the number of birth observed (Eberstadt [1994]).
8
was constantly two points above that in West Germany6. Following the fall of the Wall in
late 1989, the marriage rate dropped abruptly in East Germany in 1990 (-70 percent),
before stabilizing at around 4/1,000, meaning that the rate was similar in both regions
of re-unified Germany by the end of the period.
Combining the fertility and marriage decision information, Figure 2 displays the
difference in the yearly change in crude birth rate (per 1,000 women) between 1950 and
2008 for in and out-of wedlock birth between West and East Germany. For most of the
period, these series do not diverge by more than ten percentage points. What stands out
is that the collapse in birth rate is entirely driven by in-wedlock births (solid line), which
dropped by more than 60 percent in 1990, while those out-of wedlock (dotted line)
increased very slightly. This suggests that the cohorts born in East Germany between
1990 and 1993 were not only dramatically smaller but also negatively selected in terms
of family structure.
[Figure 2 about here]
Finally, to more precisely link the timing of the fertility drop to the regime change
in East Germany, we focus on the monthly number of births for the two regions in Figure
3. While the data for Eastern Länder is only available from January 1990 onwards, this
still enables us to observe that the number of births only started to sharply fall in August
of that year – exactly nine months after the fall of the Wall - before stabilizing in early
1994. Throughout this period, the number of births in West Germany remains
remarkably stable. The exact timing of the onset of the fall - August 1990 - is the first
indication that the collapse of the regime was not foreseen and that the drop was not
driven by immediate use of abortion (more below). As such, it was a change in the
decisions to conceive that drove the reduction in fertility. Note also that the drop in
births in the East was not solely due to displacement of mothers-to-be to the West (more
below), since the numbers of births in the West remained on a very constant trend and
pattern.
[Figure 3 about here]
6
Note that out of wedlock birth does not necessary imply single motherhood, as a large fraction of couples with
children cohabit without being formally married.
9
These figures clearly illustrate three important points that are relevant to our
identification: i) pre-1990, fertility trends were consistently similar between East and
West; ii) the fertility drop affecting East Germany after the fall of the Wall was short
lived and fertility started recovering within three and half years, as such, we define as
‘Children of the Wall’, the cohorts of individuals born between August 1990 and
December 1993 in an Eastern Länder; and iii) the marital status of parents suggests that
the CoW originated from a very different type of family.
2.3 Explaining the Fertility Drop
We consider three potential reasons why fertility fell so sharply in East Germany
after the fall of the Berlin Wall, namely change in birth control provision, East to West
migration and economic uncertainty. Although it is difficult to exactly measure the
relative importance of these factors, we provide evidence here that the reduced number
of births was mostly driven by economic considerations. While the issue of whether
women postponed or reduced their family size or whether more women remained
childless is of interest it is beyond the scope of this paper, since our aim is to understand
changes to the composition of the cohort of children born between August 1990 and
December 19937.
2.3.1 Access to Birth Control Methods
A large number of studies on fertility decisions and child outcomes have
exploited policies that changed access to birth control, and predominantly access to
abortion, which has been shown to be broadly beneficial for subsequently born cohorts
(see Bailey, Guldi and Hershbein [2013] for a brief review of this literature)8. Here,
instead, we argue that access to birth control is unlikely to be an important factor in
explaining the sudden drop in the number of births. First, access to birth control
7
However, as these delayed fertility issues could have changed the composition of individuals born after 1993,
when applicable (i.e. for analysis using SOEP data) we have tested and confirm that all our results are mostly
unchanged by the exclusion of these post treatment cohorts.
8
In particular, the US legalisation of abortion has been shown to reduce child poverty (Gruber, Levine and
Staiger, 1999); teenage motherhood (Donohue, Grogger and Levitt, 2009), use of controlled substances (Charles
and Stephens, 2006), crime (Donohue and Levitt, 2001) and improve education (Ananat, Gruber, Levine and
Staiger, 2009), while Pop-Eleches (2006) demonstrates that an abortion ban led to positive parental selection in
Romania.
10
methods was very liberal in East Germany and the right to on-demand abortion was not
modified before 1995, after which it became more restricted. Second, one could have
expected that faced with the immediate uncertainty of a new environment, potential
mothers would have terminated pregnancies in greater numbers. We have already
argued that the exact timing of the fertility drop (Figure 3) does not appear to support
this idea in the very short-run. Additionally, the number of terminations in the five East
German Länder (excluding Berlin) dropped from 72,774 in 1988 to 26,207 in 1994 (-63
percent). This more than matches the drop in the number of births observed over this
period (-57 percent), which translates into a small decrease in the abortion to birth
ratio. We can thus safely say that the decline in fertility was mostly due to a fall in
conceptions, which is important for two reasons. First, it implies that our ‘pre-treatment’
groups (of mothers and children) are not selected post-conception. Second, we can
assume that the children eventually born must have been ‘wanted’ by their mothers at
the time, which makes it a very different selection mechanism than when a drop in
fertility is driven by the legalization of abortion, and the fewer ‘unwanted’ children in a
cohort that it implies.
2.3.2 Internal Migration
One of the most important changes in the life of East Germans after the fall of the
Berlin Wall was that direct migration to the more opulent West became possible again. A
substantial number of individuals made use of this newfound freedom, with almost
800,000 individuals migrating from East to West, representing 5 percent of the pre1990 population. This internal migration flow quickly died down, and by 1993 almost as
many Germans were making the move in the opposite direction. Hunt (2006)
demonstrates that improvements in relative wages were responsible for the ebbing of
eastern migration. On average, movers were younger and more likely to be female
(Fuchs-Schündeln and Schündeln [2009]), and thus internal migration had an impact on
the reduced number of births in the East. Eberstadt (1994) estimates that internal
migration accounted for about 10 percent of the total drop in birth numbers. However,
this does not really cast doubt over the magnitude of the fertility drop, since the crude
birth rate used to illustrate it in Figures 1 and 2 uses the yearly number of women in the
population of East or West Germany as a denominator.
11
Migration remains a worry for the validity of our identification, even if it does not
directly explain the drop in fertility, since it could still distort the composition of the
(control) cohorts of individuals that we observe in West Germany. This would be the
case if mothers of young children migrated in substantial numbers or if many of the
women who moved to the West subsequently gave birth there, although this is not
observed in the raw data presented as West Germany birth numbers remain on trend.
Note that internal migration is not an issue for all of our micro-level analysis using SOEP
data, since we allocate the treatment status based on the mother’s place of residence in
1989, rather than current location. Results from the different sources point to the same
mechanism, as such internal migration is unlikely to be the driver of the observed
selection effects. Additionally, we run regressions based on the SOEP data using current
location rather than location of the mother in 1989 to allocate treatment, to test how
much internal migration may bias our results. Results from these regressions are never
statistically different from those presented. As such, we are confident that our results
are not driven by internal migration.
2.3.3 Economic Uncertainty
During the half-century of communist rule, there was no uncertainty concerning
employment and wages, and women were very integrated into the labor force. The costs
of having children were kept low due to the public provision of childcare, health and
educational services. In the months immediately following the fall of the Berlin Wall, full
employment policies were abandoned; indeed, almost a third of the pre-unification jobs
had been eliminated by the end of 1994, and 65 percent of those unemployed were
women. The generous and universal benefits linked to having a child were quickly
curtailed to match Western levels, while the availability of childcare shrank and housing
costs surged (Rheinheckel et al. [1998]). This negative economic picture was mitigated
by the aforementioned rapid catch up of Eastern wages, which were negotiated to reach
parity with the West by 1994, owing to large financial transfers from the West and a
generous one-to-one conversion of the OstMark to the DeutscheMark in July 1990. In
fact, mean disposable income and consumption in the new Länder had recovered to
their pre-1989 level as early as three years after the fall of the Wall (Dornbusch and
Wolf [1992]). Therefore, there was very substantial and fast economic convergence
between East and West Germany after the fall of the Berlin Wall, even if income
12
inequality increased in the East. Can we thus still claim that economic uncertainty drove
the drastic fall in the number of birth in those years? We use evidence from survey data
collected at the time, which links fertility decisions and uncertainty about the economic
situation to answer this question.
[Figure 4 about here]
First, the 1992 Population Policy Acceptance Study (PPAS) allows us to link the
perception of economic uncertainty to fertility decisions9. When asked in this survey
what were the reasons for not wanting a(nother) child, the most common reason given
by 78 percent of East Germans was poor economic circumstances. The next two most
common answers were also related to the perception of the economic situation, namely
the costs of raising children (60 percent) and fear of the future (49 percent).
Additionally, the SOEP allow us to track the evolution of the perception of economic
situation and childcare provision over time. Figure 4 reports the difference between
East and West Germany in terms of the fraction of individuals worried about the
economic situation. Following reunification, East Germans were 20 percentage points
more likely to be very worried about the economy. This difference increased up to 30
percentage points in 1991, before the views on the economy converged by 1993 and
remain close thereafter. Amazingly, this is precisely when we start observing a rebound
in birth rates in the East, which is consistent with our assumption that economic
uncertainty was one of the main factors behind the drop in fertility in the East10. Since
the PPAS indicates that childcare was also an important concern, the SOEP also enables
us to assess the differences in the perception of childcare availability between East and
West over time. Again, we observe that East German parents were more worried about
childcare availability, yet they rapidly converged towards the West perception. These
measures of uncertainty about the future thus validate the definition of the CoW, since
9
The Population Policy Acceptance Study (PPAS) is a comparative survey of European attitudes and opinions
concerning demographic changes, demographic behaviours and population-related policies. In Germany, the first
survey was conducted in 1992. About 10,000 men and women in East and West Germany between the ages of 20
and 39 years were asked about family policy, its impact and expectations on future family policies. For more on
this survey, see: http://www.bib-demografie.de/EN/Research/Surveys/PPAS/ppas_node.html
10
Additionally, in 1991, 45% of East German workers asked about their probability of losing their jobs within
the next 12 months reported that they would definitely or probably lose it. For East Germans, this perceived
probability of job loss fell to 21% and 16% by 1993 and 1996, respectively. Despite still being higher than in the
West, which remained between 6 and 8% during the same period, this shows a very high level of economic
uncertainty and a remarkable convergence of perceptions within the three years following re-unification.
13
the expectations about the economy and childcare of both East and West Germans had
broadly converged by 1993.
3. Data Sources and Empirical Strategy
3.1 The Datasets
3.1.1 Cross Sectional Standardized Test Data: IGLU 2001 and PISA 2006
No administrative test score data is available across states in Germany due to the
Länder’s strong independence from the central government in terms of educational
scrutiny. However, we are able to identify two international testing exercises that were
taken by large samples of German school children conceived just before and after the fall
of the Berlin Wall at various stages of their educational careers: aged 10 with the
Progress in International Reading Literacy Study (PIRLS) in 2001 and aged 15 with the
Program for International Student Assessment (PISA) in 200611.
For our analysis, we actually use an over-sample of 10,000 students of PIRLS
2001 called IGLU 2001. The questionnaire and testing are identical and the data
provider (IQB) has identified for us children attending schools in East Germany12. The
sampling included 4th grade children in 2001, 25 percent of whom were born before
August 1990. Limiting ourselves to children born in Germany between July 1989 and
June 1991 leads to 20% of pupils being defined as CoW. The test took place in May 2001
in all schools and is designed to assess the reading competences of 4th graders in
reading, comprehension and literacy13. In addition to test results, the survey collects
information from parents, which is used to create an Index of Early Home Literacy
Activities, an Index of Home Educational Resources, an Index of Parents’ Attitudes
toward Reading and a report on the number of books at home. We use these to assess
the home environment of the pupils in terms of parental input.
Similarly, PISA 2006 is an international testing exercise of 15-year-old students
(typically in grade 7) across the world. The PISA assesses the reading and math skills of
11
See Mulis et al. (2003) and OECD (2007) for general information on the PIRLS 2001 and PISA 2006 tests,
respectively.
12
This representative sample is drawn from schools in six Landër: Baden-Württemberg, Bavaria, Bremen,
Hessen and North Rhine-Westphalia all in the West and Brandenburg in the East.
13
For each competency, five plausible values reflecting the child ability are recorded. We take the average from
these fifteen plausible values as our measure of competence and normalize it to a mean of 0 and a standard
deviation of 1.
14
students, as well as collecting survey information from pupils, parents and teachers.
Typically, the testing last for about two hours per student through a combination of
multiple choice questionnaires and open-ended questions. Germany over-sampled the
2006 PISA and IQB has identified for us schools located in the former East Germany,
excluding Berlin, and we rely on school fixed effect models to eliminate any statespecific effects. The German sample contains 34,516 children, of whom we keep those
born in 1990 in Germany (30,650), 12% of whom are CoW (born in or after August 1990
and currently living in East Germany).
3.1.3 Individual Survey Data: DJI and SOEP
The DJI Youth Survey is part of the continuous social reporting undertaken at the
German Youth Institute. Here, we use the 2003 wave for youths born between 1989 and
1991, which gives us a representative sample of 2,154 German youths. These individuals
are observed when aged between 12 and 14 years and answer a battery of questions on
various topics including education and family life. It contains 7 percent of individuals
who can be classified as ‘Children of the Wall’, i.e. those identified as being born in an
East German Länder between August 1990 and December 1993.
The German Socioeconomic Panel (SOEP) is a large annual longitudinal survey of
private households first established in West Germany in 1984. Since 1990, it has also
included individuals from the former East Germany. We use data from 1990 to 2011
comprising more than 50,000 unique individuals, a quarter of who live in the East. The
SOEP includes detailed personal characteristics and extensive questionnaires for all
members of the households, including retrospective information when necessary. The
main survey is augmented by topic specific modules and we make extensive use of those
with questions focusing on mothers and young adults (aged 17) when children of the
relevant cohorts are interviewed for the first time. Note that the SOEP asks adults their
location in 1989. When using this survey, CoW is thus defined based on their 1989
location independently of future migration decisions.
In addition to basic socio-demographic characteristics, from the various
questionnaires in DJI and SOEP we extract self-reported measures of education and
15
family composition, as well as information on parenting behavior/relationship as
reported by the children14.
3.2 Empirical Strategy
For all outcomes, our empirical strategy relies on a difference-in-differences
approach whereby we compare the characteristics or educational outcome of pupils
born (conceived) before August 1990 (November 1989) to those born earlier. The
counterfactual, or second difference, is provided by the non-treated individuals from
West German Länder, which enable us to naturally control for common macro shocks
and time trends. The basic specification used throughout is as follows:
=
+
+
+
,
+
(1)
+
The subscript s denotes either a state or a school, depending on the dataset being used.
When available, a school or state fixed effect, γ, is introduced.
is a dummy for living
in East Germany, while MoB and YoB are indicators of the month and year of birth. X is a
vector of individual level characteristics, which varies between datasets. εis is an error
term assumed to be independent and normally distributed across individuals i. The
coefficient of interest in all regressions is the estimate of β on CoW, which is a dummy
equal to 1 when an individual is a Child of the Wall (i.e. born August 1990 to December
1993 in an Eastern Länder) or her mother and zero otherwise. All regressions are reweighted to account for survey design and standard errors are clustered at the school
level (IGLU and PISA) or by region and birth year (DJI and SOEP).
In a difference-in-differences framework, the identification assumption is that
there is no difference in trends between the regions before the treatment occurs. Using
SOEP, we are able to test this assumption at either the mother or child level. We never
find any statistically different pre-trend differences between East and West for the
cohorts born between 1982 and 1989 for any of the outcomes of interest.
4. Empirical Evidence on Educational Outcomes
14
Detailed information on the DJI and the SOEP is available online at: http://www.dji.de/index.php?id=1&L=1
and http://panel.SOEP.de/
16
4.1 Test Score Results
The PIRLS test assesses the reading ability of pupils in grade 4 when they are
about 10 years old. We rely on the difference in difference framework explained above,
whereby we compare test results for children from the same classroom born before and
after August 1990 in East and West Germany. Since all children are tested on the same
day, it is important to control for age at test, via month of birth dummies. We assume
that any month of birth effect on test score is similar between West and East German
schools and test this assumption below15. We additionally control for gender, number of
children in the household and dummies for whether the parents were born abroad. The
coefficient of interest is the interaction between being born after August 1990 and living
in the East, which identifies any difference in performance for the cohort of East German
children conceived after the fall of the Wall. The upper panel of Table 1 reports the
estimate of the interaction term on three outcomes: normalized test score, as well as an
indicator of being in the top or bottom of the test score distribution, respectively. CoW
score 0.15 of a standard deviation lower than their classroom peers conceived before
the fall of the Wall. This is mostly driven by the distribution of test scores for the CoW
having a larger tail of low achievers, whereby there is no effect of CoW on the probability
of being in the top decile of the distribution, although being a CoW increases the
probability of being in the bottom 10 percent by two-thirds. An effect on test score at an
early age is likely to have a large impact on educational attainment since Germany is
characterized by an early tracking system whereby pupils are streamed in grade 5 or 6,
depending on the state in which they reside.
[Table 1 around here]
We similarly analyze results of the PISA test, which assesses reading and math skills
when pupils are about 15 years old. The identification is again a difference-in15
Threats to this assumption would be that school years are organised differently in East and West German
States leading to months of birth having a different effect on grades in the two regions. Another threat would be
that variations in cohort composition across months (Buckels and Hungerman, 2013) differ between the two
regions. In an alternative specification, we include interactions between months of birth and living in the East.
The month of birth interactions become negative and significant from August onwards, and the point estimates
do not significantly differ between August and December; highlighting that over this (short) period, the selection
effect was constant.
17
differences, whereby we compare the test score of children born between January and
July 1990 to those of children born after August 1990 in the same school, in East and
West Germany. The base specification controls for month of birth, gender and whether
the parents were born abroad. The mean normalized math test score is 0.064 of a
standard deviation lower for CoW and the effects for reading are similar but larger at 0.078 of a standard deviation16. Again, we find that the results are mostly driven by a
worsening in the left tail of the test score distribution, whereby CoW are 22 (28) percent
more likely to be in the bottom decile in math (reading), while no effect is observed at
the top end of the distribution17.
To test the assumption that month of birth effects are similar in East and West
Germany, we conduct a placebo regression using the PISA 2003 where we consider the
treated as pupils born from August 1987 to December 1987 in East Germany.
Reassuringly, we do not find any effect of the placebo treatment, which assures us that
our results are not driven by region-specific month of birth effects. Finally, note that
since we control for school-specific fixed effects and compare children in the same
classroom, these results are not driven by changes to the curriculum or other
institutional differences that would affect only East German children born after August
1990. To recap, when comparing the test performances of children in the same school,
those conceived after the fall of the Wall performed substantially worse, which is
consistent with parental selection.
4.2 Self-Reported Educational Attainment
As well as those objective measures of educational performance, the DJI and
SOEP provide self-reported measures of educational attainment, with the results
presented in Table 2. Focusing on the interaction between being born post-August 1990
and being educated in the East, we find that CoW are 6.5 percentage points more likely
to have already repeated a grade. In terms of mean size impact, this effect is large and
translates into a 45 percent increase in the probability of repeat. Similarly, they were
also 40 percent less likely to report finding learning easy and 19 percent more of them
16
The results are not sensitive to using a smaller window around the Fall of the Wall. The estimates using only
children born between May and November 1990 are -0.073 (0.027) and -0.055 (0.027) for reading and math,
respectively.
17
We also obtained results for regressions that control for grade attended, track type, number of siblings, age of
mother, marital status and maternal education, although these substantially reduce the sample size (from 28,008
to 23,393 observations) as not all parents responded to the survey. The results are not significantly different for
these specifications and thus they are not presented to save space, but they are available upon request.
18
reported that they did not get on well with their peers, which are two indicators of a
lower taste for schooling.
[Table 2 around here]
Similarly, at age 17, using SOEP we find that CoW mostly display negative
educational outcomes with the results reported in the lower panel of Table 2. We
observe that they are two percentage points more likely to have dropped out of
education. This is a relatively rare outcome in Germany and thus represents a very large
mean size effect of a 53 per cent increase in the probability of not being at school at that
age. Conditional on having not dropped out, CoW are a third more likely to be in a lower
track. No effect is found on repeating, although this outcome is reported only conditional
on still being in education, and is thus a lower bound effect.
Overall, we have consistently found that the CoW display or report much worse
educational outcomes from an early age onwards, which is mostly driven by a worsening
of the tail end of the distribution. In terms of size, our effects are for example
comparable by age 10 (a -0.150 of a standard deviation in reading score) to the
difference in test scores between children who attended or not the Head-Start preschool program (taking Deming [2009]’s 0.133 estimate). The almost fifty percent
increase in drop-out rates by age 17 is very large but in line with the impact on female
high school graduation from enrolment or not in the Perry Preschool (taking Anderson
[2008]’s .494 percentage points estimate). We now explore whether this is consistent
with negative parental selection as the underlying mechanism using various measures of
this phenomenon in terms of the mother’s and family characteristics, as well as the
perceived quality of parenting and relationship as reported by the children themselves.
5. Who Gives Birth in Times of Economic Uncertainty?
5.1 Parental Selection
5.1.1 Mothers’ Socio-Economic Characteristics
19
As already discussed, the large fertility drop that we study is certainly not
random across women and is likely to be driven by parental selection. After reviewing
the evidence on the educational attainment of the CoW, our prior is that they were the
product of important negative selection into motherhood. Faced with a high level of
uncertainty about the future and a new set of (unknown) constraints regarding the costs
of child rearing, women with lower parenting skills were relatively more likely to
conceive and give birth in the years following the collapse of the Communist regime. To
test this hypothesis, we turn to the SOEP data and focus on the sub-sample of women
who gave birth in East or West Germany between 1982 and 1995. Note that the SOEP
provides retrospective information on location before the fall of the Berlin Wall, which
we use to allocate the CoW status so that these estimates are not affected by subsequent
internal migration decisions.
[Table 3 about here]
We compare the mothers of CoW to other mothers on a number of ‘positive’
socio-economic characteristics, reporting the results in Table 3. First, we note that East
German mothers are quite different to their Western peers on average over this period,
which is captured by the strongly significant coefficients on the ‘Birth East’ dummy,
although the pre-1989 trends do not differ between regions. The mothers of CoW are
over 7 months younger, almost 60 percent more likely to be teenage mothers, have nine
months less education and are eight percentage points less likely to have completed high
school. These mothers also had a lower employment probability at the time of survey.
5.1.2 Family Structure
We have already shown in Figure 2 that at the cohort level, the women who chose to
have children after the fall of the Belin Wall were much more likely to do so out of
wedlock. The DJI and SOEP allow us to assess differences in longer run family formation
much more thoroughly. As reported in Table 4, the results from the analysis of
information from both surveys reveal that the CoW experienced much less stable family
structures as they grew up. In particular, by age 12, they were 13 percent less likely to
live with their natural father, a third more likely to have experienced a divorce and had a
20
60 percent higher probability of having experienced new partnerships during their
childhood. A similar picture emerges for the family structure of these children when
they were 17 years old using the mother relationship history (since birth) in the SOEP:
by then, CoW mothers were 11 percent less likely to live with the father of the child and
had a relatively lower probability of being married. The most dramatic figure here is
that they are 80 percent less likely to have ever been married since the child was born
which is a huge effect even in view of the relatively low 6 percent average baseline.
[Table 4 around here]
The results in this section clearly confirm that our prior was correct and that women
who had children during the very uncertain times following the fall of the Berlin Wall
were negatively selected on all the standard observable socio-economic characteristics
which are associated with relatively lower educational attainment for children. While
those differences are likely to be important for child outcomes, we now investigate the
negative parental selection issue more directly, and arguably more objectively, by
relying on information on parental skills and maternal relationship, as well as parental
educational input.
5.2 Parental Input and Quality
5.2.1 Parental Inputs: Educational Inputs and Income
As well as those general indicators of the quality of the relationship, the various
surveys allow us to investigate the variations in parental inputs that are related to
education. Here, we rely on the child survey from the IGLU (age 10), DJI (age 12/13) and
PISA (age 15) to investigate differences in parental reading behavior and interest in the
child’s education. Moreover, we also assess whether one of the channels of worse
educational performance is related to deprivation.
As reported in Table 5, parents of CoW engaged in less reading activities before
the child entered school, were reading less frequently with their children at the age of 10
and their houses had less reading material. The lesser engagement of parents in the
schooling of their children is also found in the DJI, at age 12/13. Based on the child’s
answers to three questions about whether their parents care about their results, are
21
helpful in solving school problems and attend school meetings, we estimate that parents
of CoW were 0.1 of a standard deviation less engaged in the schooling of their child.
These children were also less involved in activities that could potentially compensate for
the lack of educational inputs provided by their parents. At age 15, they spent 20 less
minutes per week on their homework and were 20 percent less likely to be attending
any out-of-school teaching. An overall index of educational resources at home confirms
that they were significantly less endowed than their peers18.
[Table 5 about here]
Are these differences driven by wealth/income differentials? The last row of the PISA
panel in Table 5 suggests otherwise, revealing no significant difference in the average
wealth of CoW households as measured from children’s reporting on a list of items
available at home. To strengthen these findings, we also assess whether CoW households
are poorer, by computing the average gross and net income during childhood in the
SOEP. Consistent with the lower employment probability of mothers and their greater
propensity to live in single-headed household, CoW households have a lower gross
income. However, due to the generosity of the German welfare net, there is no
significant difference in terms of net income, meaning that the worse inputs are unlikely
to be driven by poverty. In any case, it would be unclear why – without parental
selection – economic deprivation would have disproportionally affected parents whose
children were conceived just after the fall of the Wall.
Another possible test to show that our results are not driven by income is to
include it – or other associated observable characteristics – as additional controls when
measuring the impact of being born in East Germany just after the fall on the Wall on
outcomes. Using our largest survey (PISA), we find that the estimated CoW coefficients
are slightly smaller but not statistically different when these additional controls are
included.19 This leads us to conclude that economic deprivation– or indeed, related
observable characteristics of the parents – cannot be the main driver of the markedly
worse educational outcome of their children.
18
Home educational resources is a score based on possessing the following items: A desk and quiet space to
study, a computer, education software, books to help with work, technical reference books and a dictionary. The
score is then standardized to a mean of 0 and a standard deviation of 1.
19
Our PISA results are not sensitive to including maternal education, maternal employment and family status
measures in the test regressions. The results are not reported but available on request.
22
5.2.2 Parenting Quality: Relationship and Support
The child development and psychology literature has highlighted the role of the
parental relationship and parenting style in the production of cognitive skills (see
Dornbusch et al., 1987 for example). The DJI and the SOEP provide a unique opportunity
to test usually unobservable indicators of parental skills quality. In both surveys,
children answer a battery of questions about the quality of their relationship with their
mothers, including how supportive they perceive them to be. Table 6 reports the
estimated coefficients on being a CoW using our DiD approach on these self-reported
measures of parental quality, as assessed by the child at age 12 (DJI) and 17 (SOEP).
Note that we mostly focus here on maternal relationship and support as we have
documented a large negative selectivity in the probability of these children living with
their fathers.
The DJI allows us to build a score on parental relationship quality based on the
sum of answers to the following questions: “How satisfied are you currently with your
maternal/paternal relationship?”; “Do you have a good relationship with maternal
figure?”; “Does your maternal figure support you when you need it?”; and “How
important is your maternal figure?”. This score is normalized to a mean of 0 and a
standard deviation of 1 and is scaled so that a higher value means a better quality
relationship. CoW rated their relationship with their mothers-to-be 0.1 of a standard
deviation worse than their peers. Similarly, “argument with mother” is a normalized
score based on the sum of answers to questions about “How often do you have
arguments about…” the following issues: manners, appearances, untidiness, going out,
music, political views, friends, girl/boyfriends and help with the house. A positive value
indicates a greater frequency of arguments. CoW reported a higher frequency of
arguments with their mothers, by .04 of a standard deviation, although this coefficient is
not statistically significant. Finally, the DJI allows us to construct another measure of the
child’s perceive quality of life at home, as reported in the difficult family score. This is
composed from answers to questions on a four-point scale regarding “whether there are
frictions in the family”, “whether one can speak about anything”, “whether we have fun
together” and “whether we all go our own ways”. A higher value of this normalized score
reflects a less integrated family. Again, according to our results, CoW rated their family
life much more poorly than their peers (0.2 of a standard deviation).
23
[Table 6 around here]
Similarly, at age 17, the SOEP includes a substantial number of questions on
children’s perceived quality of their maternal relationship and the support received. We
focus on three questions: “Mother Shows that she Loves You”, from which we generate a
dummy variable (“Mother Loves Me”) that takes the value 1 for answering ‘very often’
and 0 otherwise; “Fight with Mother”, which is derived from the answer to a four-point
scale question “Argue Or Fight With Mother”, from which we create a dummy variable
that takes the value 1 when the answer is ‘very often’ and 0 otherwise; and “Supportive
Parenting”, which is derived from a multi-item scale of nine questions described and
tested extensively in Weinhardt and Schupp (2011). A first strong indicator of maternal
attachment is whether teenagers feel (very often) loved by their mothers. Our estimate
in Table 6 indicates that two-thirds of CoW are less likely to be in this category,
suggesting a much lower level of maternal attachment. Another strong predictor of the
quality of the child-parent relationship is the frequency of fights, which is a relatively
rare event, with only 4 percent of children reporting this as occurring very often.
However, this probability increases by almost 150 percent for CoW, indicating a
worsening of the maternal relationship since age 12, as arguing with mother was not
significantly different for these children at that age. Finally, we use an overall measure of
‘supportive parenting’ to gauge maternal participation in the child's life and how much
the parent involved the child in decision-making. This reveals that CoW report a much
lower level of maternal support on average, at 0.3 of a standard deviation lower,
compared with other children.
These very robust findings on poorer maternal relationship quality and the low
perceived support received by CoW at different points in their childhood are important
for two reasons. First, they are perhaps surprising, given that one might have assumed
that women who had children during uncertain economic times may have wanted them
relatively ‘more’ and would have been expected to be more attached to their child later
in life. Second, they point to a potentially crucial yet often unexplored channel by which
selection into motherhood links to parental skills that drive the child’s later outcomes.
To further explore these issues, we carry out two extensions that exploit the unique
nature of the SOEP data to further test parental selection in bad economic times.
24
5.3 Testing Parental Selection in Bad Economic Times
5.3.1 Direct Evidence of Selectivity into Fertility in Bad Times
Thus far, we have provided a wealth of evidence that the women who had
children in the aftermath of the fall of the Wall where on average negatively selected,
whereby we have attributed this selection to the high level of economic uncertainty
during this period. To more directly test this mechanism, we exploit the longitudinal
information in the SOEP to combine answers for all women who answered three
relevant questions about: (i) economic uncertainty; (ii) fertility decision; and (iii)
education level. Practically, we regress the probability of having a child in the period
1991/93 on education level for all women aged 17 to 47 who were interviewed in the
SOEP, on a measure of economic uncertainty in year t-1 (i.e. dummy for being ‘very
worried’ about ‘the general economic development’).20 We find that perceived economic
uncertainty is negatively related to fertility decision in the following year for all women,
thus confirming our previous cohort level evidence. We subsequently include an
interaction of years of education and economic uncertainty in the probability model that
we estimate, whereby this interaction is negative and significant.
[Figure 5 about here]
This is best illustrated in Figure 5, which reports the estimated probability of
giving birth by education level, split by level of worry about the economy in the previous
period. A first observation is that more worried women (solid line) are less likely on
average to have a child a year later. Interestingly, at a low level of education, there is
little difference in the probability of giving birth between the very and not so worried
women. By contrast, at a higher level of education, a fertility gap opens between the two
groups, to the extent that highly educated women who are very worried about the
economy are 50% less likely to give birth in the next period compared with those of the
20
The model also includes education, age and year dummies and the standard errors are clustered at East level to
account for important common age shocks on fertility, which are likely to be different between East and West
Germany.
25
same education level who are not worried. This evidence reinforces our argument that
economic uncertainty not only affects the fertility of mothers but also their selections,
whereby those with disproportionally unfavorable characteristics are less responsive to
economic shocks.
5.3.2 Worse Mothers or Bad Times?
Finally, despite strong evidence of parental selection, the differences in the
characteristics and behavior of the CoW could also be consistent with the fetal
programming (Barker [1995]) and early life adversity (Conti and Heckman [2013])
hypotheses. For example, Aizer, Stroud and Buka (2009) show that maternal stress in
utero has long-term negative consequences for children, and that this effect is stronger
for low socio-economic status mothers. Due to the high level of uncertainty faced by
mothers after the end of Communist East Germany, these children could have
experienced heightened levels of stress in the womb and during their very early years.
In turn, this could have shaped their preferences and behavior in a way to cope with
such a world, which may have caused the lower outcomes that we have observed for
these children. Is this a credible explanation and what could we do to test for this
underlying mechanism? In any case, our previous results indicating that the mothers of
CoW had worse parental skills and that these children performed worse than their peers
who grew up in the same environment are difficult to reconcile with this theory alone.
Nonetheless, we propose two simple robustness test of the early life adversity
hypothesis.
First we run a placebo regression whereby the CoW treatment is redefined as
children born between March 1986 to July 1989 and drop all children born from August
1990 onwards. These children were conceived before any social unrest started in East
Germany but started schools in re-unified Germany. As such, they were not selected at
birth but did experience the disruption and stress of the regime change at a young age.
For this cohort, we do not observe any significant negative effects on educational
attainment or supportive parenting compared to our control groups. This confirms that
the negative effects found for the CoW are driven by parental selection and not directly
by the disruptive economic and social environment during early childhood.
26
Additionally, we provide a stronger test that the CoW effects are driven by
parental selection by using the family identifier in the SOEP, and identifying all children
born between January 1987 and July 1989 who have brothers or sisters born between
August 1989 and December 1993 in East Germany. We label these children CoW Siblings.
These children could not have been “programmed” since they were born before the
uncertainty following the collapse of the Berlin Wall and - in the absence of negative
parental selection - should not report different outcomes to other children, as seen
above. If they do, it would strongly indicate that the negative outcomes that we have
observed are due to the poorer parenting skills of the mothers they share in common
rather than because CoW were born in difficult economic times. For this sample, we
conduct an estimation akin to our general DiD approach to estimate the education and
relationship outcomes observed at age 17 in SOEP, albeit with the main coefficient of
interest now the dummy of being a CoW Sibling born before the fall of the Berlin Wall.
[Table 7 around here]
The estimates reported in Table 7 indicate that CoW Siblings also display
relatively worse outcomes on a number of our educational attainments. In particular,
they are 50% more likely to have repeated a grade by age 17 and are as likely as their
younger siblings to have dropped out (although not significantly, due to the smaller
sample size). Similarly, they report a worse quality relationship with their mothers. The
estimated coefficients on ‘mother loves me’ and ‘supportive parenting’ are very similar
for the older and younger siblings, indicating that this is likely to be a mother’s
characteristics. Since CoW siblings experienced the relative certain times of the old
regime while in the womb and during their very early life, this strongly supports the
notion that the observed effects for the CoW are due to negative parental selection and
not to fetal programming, which we thus reject as the underlying mechanism behind our
findings21.
21
An alternative way to test that the CoW effects are driven by family characteristics rather than the economic
and social environment is to run a family fixed effect model. This directly compares the educational outcome
and parenting skills of siblings, after accounting for the unobservable fixed family characteristics. If the CoW
effects are driven by parental selection, we would indeed expect that the within family estimates would be
insignificant. Indeed, we find that within sibling educational outcome differences are close to zero for CoWs.
The effects on parenting competence are also insignificant apart from fighting with mothers but are much less
precisely estimated – see table results in on-line appendix. Altogether, these tests support that the worse
outcomes observed for CoW are driven by negative parental selection rather than a time of birth effects.
27
6. Conclusion
This paper highlights that the economic environment strongly influences not only cohort
sizes but also cohort composition. Using the natural experiment created by the fall of the
Berlin Wall and the subsequent temporary collapse of fertility in East Germany, we
report that children conceived during the time of great economic uncertainty performed
worse on various dimensions of their schooling. These effects are driven by differences
in the observable characteristics of mothers, as well as by dissimilarities in behavior; for
instance, mothers who conceived in the aftermath of the fall of the Berlin Wall provided
less educational inputs to their children and had lower emotional attachments. The
differences are also observed for their older children, who were conceived at the time of
the relative economic stability associated with the communist regime, thus highlighting
that the results are driven by parental selection and not a specific time of birth effect.
Our findings concerning the large effects of parental selections on the outcomes
of future generations have important implications for policy planners. First, rather than
basing decisions regarding public investment on cohort size only, there is scope for
adjusting these investments for cohort quality, especially if peer effects are important. In
this case, despite its small size, this cohort would have benefited from additional
investment to compensate for the lower parental provision. However, divergence in
educational outcome starts early, meaning that any interventions to compensate for the
worse parental skills would have to take place early in childhood (Cunha and Heckman
[2007]) and focus on affecting personality skills (Heckman, Pinto and Saveleyev [2013]).
There is however scope to try and cancel out the negative impact of parental selection
on children educational outcome with pre-school programs since the positive longer
term impact of such interventions have been estimated to be almost symmetrically
equivalent (Anderson [2008] and Deming [2009]).
Our findings also suggest that certain women will always choose to have children,
even if the conditions for making this decision are less than optimal. It is therefore
probably not possible to design policies to change their fertility behavior but however
perhaps possible to intervene at this very early stage. Experimental evidence on the
impact of home visiting programs aimed at at-risk mothers and their family that start
28
even before the birth of the child, such as Preparing for Live in Dublin (Doyle et al
[2013]), Pro Kind in Germany (Sandner [2012]) and Healthy Families America (LeCroy
and Crysik [2011]), are promising. The real challenge remains to find a way to efficiently
target such interventions at the right mothers/children since the selection effects are
driven by typically unobservable characteristics like emotional attachment and parental
educational input. Additionally, timing of birth within the business cycle could be used
as a new additional indicator to identify mothers and target children from cohorts as
being at greatest risk from poor educational outcomes.
29
References
Aizer, Anna, Laura Stroud, and Stephen Buka, “Maternal Stress and Child Outcomes:
Evidence from Siblings”, NBER Working Paper No. w18422, (2009)
Alesina, Alberto, and Nicola Fuchs-Schündeln, “Goodbye Lenin (or Not?): The Effect of
Communism on People’s Preferences,” American Economic Review, 97 (2007),
1507-1528.
Ananat, Elisabeth, Jonathan Gruber, Phillip B. Levine and Douglas Stainer. “Abortion and
Selection”, Review of Economics and Statistics, 91 (2009), 124-136
Anderson, Michael L. “Multiple Inference and Gender Differences in the Effect of Early
Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and Early
Training Project”, Journal of the American Statistical Association, 103 (2009),
1481-95
Bailey, Martha J., Melanie Guldi and Brad J. Hershbein “Recent Evidence on the Broad
Benefits of Reproductive Health Policy,” Journal of Policy Analysis and
Management, 32 (2013), 888–896.
Barker, David J.P., “Fetal Origins of Coronary Heart Disease,” British Medical Journal, 311
(1995), 171-174.
Becker, Gary S., “An Economic Analysis of Fertility” In: Demographic and Economic
Change in Developed Countries. Universities-National Bureau Series 11, Princeton
NJ: Princeton University Press (1960) 209-240
Ben-Porah, Yoram, “An Economic Analysis of Fertility in Israel: Point and Counterpoint,”
Journal of Political Economy, 81 (1973), S202-S233
Brooks-Gunn, Jeanne, Lisa, G. Berlin and Allison S. Fuligni. “Early childhood intervention
programs: What about the family?” in Jack P. Shonkoff and Samuel J. Meisels,
(Eds.), Handbook of early childhood intervention (3rd ed.) (2010). New York:
Cambridge University Press.
Buckels, Kasey S. and Daniel M. Hungerman. “Season of Birth and Later Outcomes: Old
Questions, New Answers”. Review of Economics and Statistics, 95(2013), 711-724.
Burchardi, Konrad and Tarek Hassan, “The Economic Impact of Social Ties: Evidence
from German Reunification,” Quarterly Journal of Economics, 128 (2013), 12191271.
30
Burstyn, Leonardo and Davide Cantoni, “A Tear in the Iron Curtain: The Impact of
Western Television on Consumption Behavior,” CEPR Discussion Paper No 201219, 2012.
Charles, Kerwin Kofi, and Melvin Stephens, “Abortion legalization and Adolescent
Substance use,” Journal of Law and Economics, 49 (2006), 481-505.
Conrad, Christof, Michael Lechner, and Welf Werner, “East German fertility after
unification: crisis or adaptation?” Population and Development Review, 22 (1996),
331- 358.
Conti, Gabriella and James J. Heckman, “The Economics of Child Well-Being” in The
Handbook of Child Well-Being, in Ben-Arieh, A., I. Frones, F. Casas, and J. E. Corbin
Eds. (2013), 363-401.
Cunha, Flavio, and James, J. Heckman, “The Technology of Skill Formation,” American
Economic Review, 97 (2007), 31-47.
Cunha, Flavio, James, J. Heckman, Lance Lochner and Dimitriy V. Masterov. “Interpreting
the Evidence on Life Cycle Skill Formation” in Handbook of the Economics of
Education, Vol 1, Eric Hanushek and Finis Welch (eds), Elsevier (2006).
Currie, Janet, Valentina Duque, and Irwin Garfinkel “The Great Recession and Mother’s
Health,” Economic Journal, forthcoming
Dehejia, Rajeev, and Adriana Lleras-Muney, “Booms, Busts and Babies’ Health,” Quarterly
Journal of Economics, 119 (2004), 1091-1130.
Del Bono, Emilia, Andrea Weber and Rudolf Winter-Ebmer,”Clash of Career and Family:
Fertility Decisions after Job Displacement”, Journal of the European Economic
Association, 10 (2012), 659-683.
Deming, David, “Early Childhood Intervention and Life Cycle Skill Development:
Evidence from Head Start”, American Economic Journal, Applied Economics, 1
(2009), 111-134.
Donohue, John J. III, Jeffrey Grogger, and Steven D. Levitt “The Impact of Legalized
Abortion on Teen Childbearing”, American Law and Economics Review, 11 (2009),
24-46.
Donohue, John J. III, and Steven D. Levitt, “The impact of legalized abortion on crime,”
Quarterly Journal of Economics, 116 (2001), 379-420.
31
Dornbusch, Rudiger and Holger C. Wolf, “East German Economic Reconstruction,” in The
Transition in Eastern Europe, Volume 1, Olivier Blanchard, Kenneth A. Froot, and
Jeffrey D. Sachs ed. (Chicago IL: University of Chicago Press, 1992).
Dornbusch, Sanford M., Philip L. Ritter, P. Herbert Leiderman, Donald F. Roberts, and
Michael J. Fraleigh. "The relation of parenting style to adolescent school
performance." Child development (1987): 1244-1257.
Doyle, Orla, Colm Harmon, James J. Heckman, Caitriona Logue, and Seong Moon,
“Measuring the Investment in Human Capital Formation: An Experimental
Analysis of Early Life Outcomes”, NBER Working Paper No. w19316, 2013.
Eberstadt, Nicholas, “Demographic Shocks After Communism: Eastern Germany 19891993,” Population and Development Review, 20 (1994), 137-152.
Fuchs-Schündeln, Nicola, “The Response of Household Saving to the Large Shock of
German Reunification,” American Economic Review, 98 (2008)., 1798-1828.
Fuchs-Schündeln, Nicola, and Matthias Schündeln, “Precautionary Savings and SelfSelection: Evidence for the German Reunification ‘Experiment’,” Quarterly Journal
of Economics, 120 (2005), 1085-1120.
Fuchs-Schündeln, Nicola, and Matthias Schündeln, ”Who stays, who goes, who returns?
East-West Migration within Germany since Reunification,” Economics of
Transition, 17 (2009) 703-738.
Gronau, Reuben “Leisure, Home Production, and Work-The Theory of the Allocation of
Time Revisited,” Journal of Political Economy, 85 (1977), 1099-1123
Gruber, Jonathan, Phillip B. Levine, and Douglas Staiger, “Abortion Legalization and Child
Living Circumstances: Who Is the ‘Marginal Child’,” Quarterly Journal of
Economics, 114 (1999), 263–292.
Heckman, James, Rodrigo Pinto and Peter Savelyev, “Understanding the Mechanisms
Through Which an Influential Early Childhood Program Boosted Adult
Outcomes”, American Economic Review, 103 (2013), 2052-2086.
Hunt, Jennifer, “Staunching Emigration from East Germany: Age and the Determinants of
Migration,” Journal of the European Economic Association, 4 (2006), 1014-1037.
Judt, Tony, Postwar: A History of Europe since 1945, (London, UK: Pimlico).
Kempkes, Gerhard “Rapid Demographic Change and the Allocation of Public Education
Resources: Evidence from East Germany”, Deutsche Bundesbank Discussion
Paper No. 16/2010, 2010.
32
LeCroy, C. W., & Krysik, J.” Randomized trial of the Healthy Families Arizona home
visiting program”. Children and Youth Services Review, 33 (2011), 1761-1766.
Lindo, Jason, “Are Children Really Inferior Goods? Evidence from Displacement-Driven
Income Shocks,” Journal of Human Resources”,45 (2010) 301-327
Mullis, Ina, Michael Martin, Eugenio Gonzalez, and Ann Kennedy, “PIRLS 2001,
International Report”, IEA, Boston College. (2003)
Neal, Derek “The Measured Black-White Wage Gap among Women is Too Small,” Journal
of Political Economy, 122 (2004), S1-S28
OECD “PISA 2006 Science Competencies for Tomorrow's World”, Paris, 2007.
Perry, Cynthia, “How do Female Earnings Affect Fertility Decisions?” Massachusetts
Institute of Technology, PhD Thesis, (2004)
Pop-Eleches, Christian, “The impact of an abortion ban on socioeconomic outcomes of
children: Evidence from Romania,” Journal of Political Economy, 114 (2006), 744773.
Reinheckel, Antje, Kornelia Frank, Wolfgang Weise, and Bernt-Peter Robra, “Effect of ReUnification on Fertility Behaviour in East Germany: A Review of the Evidence
Reproductive,” Health Matters, 6 (1998), 122-128.
Sandner, Malte, “Effects of Early Child Intervention on Child Development and Early Skill
Formation. Evidence from a Randomized Controlled Trial,” Hannover University
Discussion Paper No 518, (2012)
Schaller, Jessamyn “Booms, Busts and Fertility. Testing the Becker Model Using Gender
Specific Labor Demand”, University of California, Davis, Mimeo (2012)
United Nations Economic Commission for Europe, “Fertility Decline in the Transition
Economics, 1989-1998: Economic and Social Factors Revisited,” Economic Survey
of Europe, 1 (2000), 189-207.
van den Bergh, Bea, Eduard Mulder, Maarten Mennes and Vivette Glover, “Antenatal
Maternal Anxiety and Stress and the Neurobehavioural Development of the
Foetus and Child: Links and Possible Mechanisms. A review,” Neuroscience and
Biobehavioral Reviews, 29 (2005) 237–258.
Weinhardt, Michael and Jürgen Schupp, “Multi-Itemskalen im SOEP Jugendfragebogen,”
DIW Technical Report 60/2011, (2011)
33
Table 1: Test scores and related outcomes at age 10 and 15
CoW
Mean
Effect
size
Obs,
Normalized reading score
-0.150**
(0.074)
n.a.
n.a.
5,036
Overall reading score <p(10)
0.056**
(0.024)
0.085
0.660
5,036
Overall reading score > p(90)
0.002
(0.021)
0.104
0.000
5,036
At age 10- IGLU 2001
At age 15- PISA 2006
Norm. Math score
-0.064***
(0.022)
n.a.
n.a.
28,008
Math score <p(10)
0.023***
(0.009)
0.083
0.280
28,008
Math score > p(90)
0.008
(0.010)
0.105
0.070
28,008
Norm. Reading score
-0.078***
(0.020)
n.a.
n.a.
28,008
Read score <p(10)
0.017**
(0.008)
0.075
0.224
28,008
Read score > p(90)
0.002
(0.010)
0.105
0.022
28,008
Notes: CoW is defined as respondents born from August 1990 to December 1990 and schooled in East
Germany. Estimates are weighted to account for sample design and non-response. Standard errors are
clustered at the school level.
IGLU Control: Gender, mother born abroad, father born abroad, number of children in household.
month and year of birth dummies, post-August 1990 birth and a school fixed effect.
PISA Control: Gender, mother born abroad, father born abroad, month of birth dummies (all children
are born in 1990), post-August 1990 birth and a school fixed effect.
34
Table 2 – Self-Reported School Outcomes of the Children of the Wall at Ages 12 and 17
Age 12 – DJI
Age 17 - SOEP
Repeated
Grade
Learning
Easy
Gets on Well
with Peers
Low
Track
Repeated
Grade
School
Drop-Out
0.065***
(0.008)
-0.066**
(0.027)
-0.125**
(0.048)
0.027*
(0.014)
0.022
(0.024)
0.018***
(0.005)
Born East
-0.020
(0.013)
0.002
(0.012)
0.008
(0.037)
-0.050***
(0.011)
-0.036***
(0.004)
-0.001
(0.003)
Born Aug 1990-Dec 1993
0.010
(0.011)
-0.004
(0.018)
0.131***
(0.028)
-0.054***
(0.015)
-0.011
(0.017)
-0.034***
(0.006)
Yes
Yes
Yes
Yes
Yes
Yes
Mean value of outcome
0.147
0.163
0.654
0.084
0.211
0.036
Effect size at mean
0.449
-0.402
-0.190
0.321
0.103
-0.498
Sample Size
1,450
1,451
1,451
3,506
3,497
3,636
Child of the Wall
Controls
Note: CoW is an interaction of living in East Germany and being born between August 1990 and 1991 for the DJI and an indicator of being born in East Germany
between August 199 and 1993 for the SOEP. DJI estimates are re-weighted to account for design and non-response. Robust standard errors clustered by child year
of birth and East/West reported in parenthesis. *, **, and *** denote significance at the 1, 5, and 10 percent level, respectively.
DJI: Repeat grade is a dummy variable taking the value 1 if pupil reports to have already repeated a grade. ‘Learning easy’ and “Get on Well with Peers” are
dummy variables taking the value 1 if the pupil complete agrees (on a four-point scale) to these questions. The controls used in all DJI specifications include
gender, gender, mother’s age, number of siblings, as well as year and month of birth dummies.
SOEP: All information is taken from questions asked to individuals aged 17 between 1990 and 2012 (i.e. born 1982 to 1995). ‘Low Track’ indicates that the
individual reports being enrolled in the lowest educational track of the German school system (i.e. Hauptschule). The controls used in all SOEP specifications
include controls for gender, mother’s age, number of siblings, birth order, as well as year and month of birth dummies.
35
Table 3 – Positive or Negative Selection?
Differences in Characteristics of Mothers of the ‘Children of the Wall’
Age
Mother
Teenage
Mother
Years of
Education
High
School
Employed
Child of the Wall
(East * 1991-93)
-0.638***
(0.218)
0.034**
(0.015)
-0.715***
(0.125)
-0.078***
(0.019)
-0.116***
(0.027)
Birth East
-2.858***
(0.063)
0.064***
(0.008)
0.886***
(0.055)
0.135***
(0.010)
0.039***
(0.012)
Birth August 1990-93
1.257***
(0.088)
-0.040***
(0.013)
0.335***
(0.036)
0.015
(0.012)
0.046**
(0.021)
No
No
Yes
Yes
Yes
Mean value of outcome
26.47
0.058
12.258
0.872
0.774
Effect size at mean
-0.024
0.586
-0.058
-0.088
-0.150
Sample Size
4,420
4,420
4,358
4,420
4,420
SOEP
Age of Mothers
Note: Based on SOEP waves 1990 to 2012 for all women who had a child in East or West Germany between 1982 and 1995. All
specifications include number of children and child year of birth dummies. Robust standard errors clustered by child year of birth and region
reported in parenthesis. *, **, and *** denote significance at the 1, 5, and 10 percent level, respectively.
36
Table 4 – Family Composition at Age 12 (DJI) and Age 17 (SOEP)
At age 12 (DJI)
As Reported by the Child
At age 17 (SOEP)
As Reported by the Mother
Lives with
Father
Experienced
Divorce/
Separation
Experienced
New
Partnership
Still with
Father
Married
Now
Never
Married
Child of the Wall
(i.e. East * 1991-93)
-0.100***
(0.008)
0.064**
(0.020)
0.090*
(0.040)
-0.067***
(0.017)
-0.033**
(0.013)
0.046***
(0.011)
Born East
-0.106***
(0.010)
0.059***
(0.008)
0.057***
(0.006)
-0.011
(0.015)
-0.058***
(0.011)
-0.015**
(0.006)
0.035
(0.030)
-0.084*
(0.040)
-0.061
(0.018)
-0.159***
(0.007)
-0.013**
(0.005)
-0.037**
(0.010)
Mean value of outcome
0.780
0.191
0.153
0.618
0.721
0.059
Effect size at mean
-0.129
0.334
0.589
-0.109
-0.046
0.792
Sample Size
1,445
1,441
1,441
4,420
4,420
4,420
Born 1991-1993
Note: CoW is the interaction of being born between August 1990 and 1991 (DJI) or being born between 1991 and 1993 (SOEP), and living in East
Germany. Robust standard errors clustered by child year of birth and East/West reported in parenthesis. *, **, and *** denote significance at the 1,
5, and 10 percent level, respectively. Effect size is measured as the effect of CoW at the mean value for the variable: this is not reported for
normalized scores. DJI: Additional controls include year and month of birth, gender, age of mother and number of siblings. The variables of
interest are defined as follows: Experienced data: Positive answer to “Have you experienced the following event …?”; Difficult family,
normalized score from the sum of answers to the following questions “I’m happy with my family”, “our family argues”, “we can speak about
anything”, “Everyone can do what they want”, “we have fun together”. SOEP: Based on 1990 to 2012 waves for all women who had a child in
East or West Germany between 1982 and 1995. All specifications include number of children and child year of birth dummies.
37
Table 5: Parental Education Inputs and Income when Child is Aged 10 to 17
CoW
Mean
Effect
Size
Obs,
Cond.
Track
n.a.
n.a.
4,976
No
n.a
n.a.
5,220
No
83.12
-0.201
5,765
No
n.a
n.a.
1,446
No
8.134
-0.04
27,126
Yes
0.352
-0.194
28,008
Yes
n.a
n.a.
27,968
Yes
n.a
n.a.
27,997
Yes
n.a
n.a.
4,420
No
n.a
n.a.
4,420
No
At age 10 IGLU
Pre-school reading activity
Parent reading score
Number of books at home
-0.186***
(0.063)
-0.164***
(0.070)
-16.744***
(5.596)
At age 12: DJI
Parents care about school
-0.108***
(0.040)
At age 15: PISA
Homework hours
Courses outside school
Education Resources
Wealth
-0.298*
(0.167)
-0.068***
(0.016)
-0.066**
(0.031)
0.005
(0.031)
At age 17: SOEP
Household Raw Income
Household Net Income
-0.153**
(0.058)
-0.071
(0.058)
Note: CoW is defined as respondent born from August 1990 to December 1990 and schooled in East
Germany – For SOEP, the location is based on maternal residence in 1989. Estimates are weighted to account
for sample design and non-response. Standard errors are clustered at the month/year * region level. In all
surveys, controls are gender, age of mother, month of birth dummies and post-August 1990 birth. In IGLU,
number of siblings and school fixed effects are also included; in DJI, number of siblings is also included, in
PISA, dummies for parents born abroad and school fixed effects are included; IGLU: Pre-school reading is a
normalized score of activities that parents engaged in (often, sometimes, never) before the child entered
school. The activities are read, tell stories, sing, play with alphabet toys, reading games on computer, word
games, write letters, read signs, watch programs teaching how to read. Parent reading score is a normalized
score of answers (Every day, Once a week, Once a month, Never) to “How often read aloud to child?, How
often listen to child read aloud?”. Number of books is the average of the child and parents’ report on the
number of books at home. DJI: Parent care about schooling is a normalized score of answers on a four-point
scale to the question “How important is your school performance to your parents, my parents support me
with problems at school, my parents attend school meetings”. PISA: “home work hours” is the sum of the
self-reported amount of time spent studying for Science, Math, German and other subjects. “Courses outside
school” is an indicator of whether the pupil has additional courses on subject also studied at school. “Wealth”
is a normalized score based on answers to the following “have a desk”, “own room”, “a quiet place to study”,
“a computer”, “internet link”, “DVD player”, “Dish-washer” SOEP: Taken from reported income from 1990
and 2010 by women who had a child between 1982 and 1995. Net income is reported income after
accounting for social transfers. All specifications include number of children and child year of birth
dummies.
38
Table 6 – Maternal Relationship and Support as Reported by Children at age 12 and 17
Age 12 (DJI)
Age 17 (SOEP)
Relationship
with Mother
Arguments
with Mother
Difficult
Family Index
Mother
Loves Me
Fight with
Mother
Supportive
Mother Index
-0.110***
(0.015)
0.039
(0.032)
0.206***
(0.036)
-0.173**
(0.063)
0.052***
(0.016)
-0.304***
(0.094)
0.120***
(0.007)
0.222***
(0.025)
-0.167***
(0.024)
-0.076
(0.110)
-0.062
(0.037)
-0.263***
(0.028)
0.031*
(0.017)
-0.022
(0.039)
-0.033***
(0.004)
-0.011
(0.017)
0.138**
(0.059)
-0.443***
(0.156)
Mean value of outcome
n.a.
n.a.
n.a.
0.460
0.036
n.a.
Effect size at mean
n.a.
n.a.
n.a.
-0.368
1.436
n.a.
1,402
1,404
1,427
3,477
3,496
3,413
Child of the Wall
(i.e. East * 1991-93)
Born East
Born 1991-1993
Sample Size
Note: CoW is the interaction of being born between August 1990 and 1991 (DJI) or being born between August 1990 and December 1993 (SOEP),
and living in East Germany. Robust standard errors clustered by child year of birth and East/West reported in parenthesis. *, **, and *** denote
significance at the 1, 5, and 10 percent level, respectively. Effect size is measured as the effect of CoW at the mean value for the variable: this is not
reported for normalized scores. DJI additional controls include year and month of birth, gender, number of siblings and mother’s age. The variables of
interest are defined as follows: relationship with mother/father: normalized score based on the sum of answers to the following questions: “How
satisfied are you currently with your maternal/paternal relationship”, “Do you have a good relationship with maternal/paternal figure?”, “Does your
maternal/paternal figure support you when you need it?” “How important is your maternal/paternal figure?” Argument with mother/father: normalized
score based on the sum of answers to questions about “How often do you have arguments about…” manners, appearances, untidiness, going out,
music, political views, friends, girl/boyfriends, help with house. SOEP: All specifications include controls for gender, mother’s age, number of
siblings, birth order, year and month of birth. The variables of interest are defined as follows: Mother Loves Me comes from the question “Mother
Shows that she Loves You”, from which we generate a dummy variable that takes the value 1 answer is ‘very often’ or ‘often’ and 0 otherwise. Fight
with Mother derives from the answer to a four-point scale question “Argue Or Fight With Mother”, from which we create a dummy variable that takes
the value 1 when the answer is ‘very often’ and 0 otherwise. Supportive Parenting is derived from a multi-item scale of nine questions as described in
Weinhardt and Schupp (2011).
39
Table 7: Educational Attainment and Maternal Relationship at Age 17 of CoW Siblings
Educational Outcome
Maternal Relationship
SOEP – Age 17
Low
Track
Repeated
Grade
School
Drop-Out
Mother
Loves Me
Fight with
Mother
Supportive
Mother Index
Sibling of a CoW
(CoW * Born 1987 to 1989)
-0.037
(0.028)
0.121***
(0.037)
0.027
(0.017)
-0.134***
(0.039)
-0.021
(0.014)
-0.250**
(0.116)
-0.035**
(0.012)
-0.051**
(0.014)
-0.001
(0.006)
0.051**
(0.020)
-0.036***
(0.004)
0.212
(0.063)
0.024
(0.015)
-0.028
(0.025)
-0.017**
(0.007)
0.064**
(0.029)
0.009
(0.017)
0.246**
(0.112)
Yes
Yes
Yes
Yes
Yes
Yes
Mean value of outcome
0.093
0.223
0.037
0.427
0.040
-
Effect size at mean
-0.404
0.544
-0.724
-0.314
-0.515
-
Sample Size
1,995
1,988
2,072
1,944
1,953
1,906
Born East
Sibling Born
Aug 1990 – Dec 1993
Controls
Note: CoW Sibling is an indicator of being born between January 1987 and July 1990 and having a brother or sister born in East Germany between August
1990 and December 1993 (i.e. a CoW). All specifications and definitions of outcome variables are as in Table 2 and 6 above for SOEP results. Robust
standard errors clustered by child year of birth and East/West are reported in parenthesis. *, **, and *** denote significance at the 1, 5, and 10 percent level,
respectively.
40
Figure 1: Birth and Marriage Rates in East and West Germany from 1950 to 2008
20
5
4
6
10
8
15
10
12
A] Annual Crude Birth Rate per 1,000 Women from 1950 and 2008
1950
1960
1970
1980
Year
East (Left Axis)
1990
2000
2010
West (Right Axis)
4
6
8
10
12
B] Annual Marriage Rate per 1,000 Inhabitants from 1950 to 2008
1950
1960
1970
1980
Year
East
1990
2000
2010
West
Notes: Authors’ own calculations based on administrative population data from the Federal
Institute for Population Research (http://www.bib-demografie.de). East refers to the former East
Germany Länders and West to the territories of the formal Federal Republic. Berlin is omitted.
41
DiD of In and Out of Wedlock Birth Rates
-.6
-.4
-.2
0
.2
Figure 2: Birth Rate by Marital Status (In and Out of Wedlock)
Year-on-Year Difference between East and West Germany from 1950 to 2008
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Year
In Wedlock Birth
Out Wedlock Birth
Notes: Graph shows the differences-in-differences coefficients of the change in the year-on-year
birth rate by marital status between East and West Germany. Authors’ own calculations based on
administrative population data from the Federal Institute for Population Research.
42
65000
50000
August 1990 (Fall of Wall + 9 Months)
5000
35000
10000
15000
Figure 3: Monthly Number of Births in East and West Germany from 1990 to 2000
1990m1
1992m1
1994m1
1996m1
Month of Birth
East (Left Axis)
1998m1
2000m1
West (Right Axis)
Notes: Administrative birth data from the Federal Institute for Population Research
43
0
Difference East - West
.1
.2
.3
Figure 4: Difference in the Proportion of East and West Germans
who are Very Worried about the Economy or Childcare from 1990 to 1996
1990
1991
1992
1993
Year
Economy
1994
1995
1996
Childcare
Note: The graphs are based on the difference in the proportion of East and West Germans responding ‘very’
(other possible answers: ‘somewhat’ or ‘not at all’) to questions asked yearly in the SOEP concerning
individual level of worry about “the general economic development” and “childcare availability”.
44
.01
Probability of Having a Child
.02
.03
.04
.05
.06
Figure 5: Economic Uncertainty and Fertility Decision:
Probability of Having a Child by Economic Worry and Education Level
7
8
9
10
11
12
13
14
Years of Education
Very Worried about the Economy
15
16
17
18
Not Very Worried
Note: The graph plots the estimated probability of having a child in the period 1991/93 separately for individuals
reported to be very worried about the economy (‘very’ = 1 and ‘somewhat’/‘never = 0) or not, by years of
education for all women aged 17 to 47 surveyed in SOEP during this period. The probit model that generates
these coefficients also includes education, age and year dummies. The gray area represents the 95 percent
confidence intervals.
45
Online Appendix Tables with Additional Results – Not for Publication
Table A1: Test Scores at age 15 – Placebo Cohort (i.e. Using PISA 2003 and Treatment:
Born from August 1987 to December 1987 and Schooled in East Germany)
CoW
Mean
Effect
size
Obs,
Norm. Math score
-0.023
(0.020)
n.a.
n.a.
31,716
Math score <p(10)
-0.003
(0.008)
0.099
-0.034
31,716
Math score > p(90)
0.008
(0.008)
0.100
0.085
31,716
Norm. Reading score
-0.031
(0.022)
n.a.
n.a.
31,716
Read score <p(10)
0.014*
(0.008)
0.100
0.139
31,716
Read score > p(90)
0.025***
(0.009)
0.100
0.251
31,716
Placebo Treatment
At age 15- PISA 2003
Notes: Estimates are weighted to account for sample design and non-response. Standard errors are
clustered at the school level. PISA Control: Gender, mother born abroad, father born abroad, month of
birth dummies (all children are born in 1990), post-August 1990 birth and a school fixed effect.
46
Table A2 – Education and Support at Age 17 for Placebo Cohort (Treatment: Born East in 3 Years and 5 Months before Fall of Wall)
Educational Outcome
Maternal Relationship
SOEP – Age 17
Low
Track
Repeated
Grade
School
Drop-Out
Mother
Loves Me
Fight with
Mother
Supportive
Mother Index
Born East * Born March
1986 to July 1989
-0.022
(0.016)
-0.000
(0.016)
-0.003
(0.006)
-0.016
(0.026)
-0.004
(0.008)
-0.071
(0.129)
Born East
-0.040***
(0.011)
-0.054***
(0.012)
0.001
(0.005)
0.042*
(0.021)
-0.032***
(0.004)
-0.162
(0.053)
Born March 1986 to July
1989
0.082***
(0.012)
0.021
(0.025)
-0.005
(0.012)
0.018
(0.052)
0.052***
(0.013)
0.369**
(0.128)
Yes
Yes
Yes
Yes
Yes
Yes
Mean value of outcome
0.085
0.214
0.037
0.456
0.036
-
Effect size at mean
-0.259
-0.001
-0.072
-0.034
-0.106
-
Sample Size
2,729
2,721
2,834
2,696
2,696
2,629
Controls
Note: Robust standard errors clustered by child year of birth and East/West reported in parenthesis. *, **, and *** denote significance at the 1, 5, and 10
percent level, respectively. Effect size is measured as the effect of CoW at the mean value for the variable: this is not reported for normalized scores.
All specifications include controls for gender, mother’s age, number of siblings, birth order, year and month of birth. The variables of interest are defined as
follows: ‘Low Track’ indicates that the individual reports being enrolled in the lowest educational track of the German school system (i.e. Hauptschule).
Mother Loves Me comes from the question “Mother Shows that she Loves You”, from which we generate a dummy variable that takes the value 1 answer is
‘very often’ or ‘often’ and 0 otherwise. Fight with Mother derives from the answer to a four-point scale question “Argue Or Fight With Mother”, from which
we create a dummy variable that takes the value 1 when the answer is ‘very often’ and 0 otherwise. Supportive Parenting is derived from a multi-item scale
of nine questions as described in Weinhardt and Schupp (2011).
47
Table A3 – Education and Support at age 17 – Family Fixed Effects Evidence
Educational Outcome
Maternal Relationship
SOEP – Age 17
Low
Track
Repeated
Grade
School
Drop-Out
Mother
Loves Me
Fight with
Mother
Supportive
Mother Index
East * Born August 1990 to
December 1993
0.072
(0.045)
-0.005
(0.087)
0.007
(0.034)
-0.099
(0.106)
0.106**
(0.043)
0.222
(0.352)
Born East
-0.012
(0.165)
0.207
(0.185)
-0.028
(0.030)
0.115
(0.239)
-0.182
(0.115)
-0.280
(1.170)
Born August 1990 to
December 1993
-0.113*
(0.057)
-0.013
(0.065)
-0.045
(0.043)
0.095
(0.082)
-0.038
(0.037)
0.203
(0.464)
Yes
Yes
Yes
Yes
Yes
Yes
Mean value of outcome
0.084
0.211
0.036
0.460
0.036
-
Effect size at mean
0.854
-0.024
0.184
-0.215
2.932
-
Sample Size
3,506
3,497
3,636
3,477
3,496
3,413
Controls
Note: Robust standard errors clustered by child year of birth and East/West reported in parenthesis. *, **, and *** denote significance at the 1, 5, and 10
percent level, respectively. Effect size is measured as the effect of CoW at the mean value for the variable: this is not reported for normalized scores.
All specifications include controls for gender, mother’s age, number of siblings, birth order, year and month of birth. The variables of interest are defined as
follows: ‘Low Track’ indicates that the individual reports being enrolled in the lowest educational track of the German school system (i.e. Hauptschule).
Mother Loves Me comes from the question “Mother Shows that she Loves You”, from which we generate a dummy variable that takes the value 1 answer is
‘very often’ or ‘often’ and 0 otherwise. Fight with Mother derives from the answer to a four-point scale question “Argue Or Fight With Mother”, from which
we create a dummy variable that takes the value 1 when the answer is ‘very often’ and 0 otherwise. Supportive Parenting is derived from a multi-item scale
of nine questions as described in Weinhardt and Schupp (2011).
48